Streamflow forecasting by modeling the rainfall-streamflow relationship using artificial neural networks

Streamflow forecasting is a complex and fundamental hydrological phenomenon. The accurate prediction of the streamflow helps in the planning, design, and management of water resources in particular irrigation, hydropower production, flood risk management, and protection for dams. Unexpected and heavy rainfall results in river overflowing which is the leading cause of serious flooding. In this research work, we proposed an effective approach for streamflow forecasting by modeling the rainfall-streamflow relationship. The key objective of this research work is to identify the appropriate set of rainfall patterns to predict the daily river streamflow. This research work is divided into two successive phases. In the first phase of this research, we have identified the different sets of antecedent rainfall combinations. In the second phase, the artificial neural network (ANN) models are developed and trained using each of these different rainfall patterns to forecast daily river streamflow. Finally, the performance of the developed ANN models is evaluated using four different performance metrics including root-mean-squared error (RMSE), correlation coefficient (R), the coefficient of determination (R-2), and Nash-Sutcliffe efficiency coefficient. The results of the research work indicate that the ANN model developed by presenting rainfall patterns of the previous 4 days can precisely predict the daily streamflow with 0.97 and 0.94 value of R-2 for the validation and test period, respectively. The results of the research work also revealed that the architecture of the ANN model and combination of input patterns presented to the model significantly affects the training time, learning ability, and performance of the ANN model. Furthermore, these encouraging results proved that the proposed ANN-based approach can be a useful and effective alternative method for solving hydrological problems.